Impressive, but not an AI PC: The AI kit for the Raspberry Pi, which calculates with 13 TOPS, is economical but not exactly cheap. Hobbyists have been using AI on the Raspberry Pi for a long time, for example, for image recognition (g+) or to combat loud neighbors. But the Llama language model also runs on the minicomputer. However, the limited computing power quickly thwarts your projects, not to mention their efficiency. External AI accelerators like the AI Kit that Raspberry Pi Limited introduced make this better.
The kit consists of the M.2 Hat+, which was also recently introduced, and a matching circuit board with an AI accelerator. Like the entire circuit board, it comes from the Israeli developer Hailo, not to be confused with the conductor manufacturer of the same name. The built-in Hailo 8L chip calculates at 13 TOPS, significantly faster than many other AI accelerators for minicomputers, and is very economical with a maximum power consumption of 2.3 watts.
This makes the official kit significantly more powerful than Google’s Edge TPU, aka Coral, which was previously popular as an AI accelerator. Since the connection is via PCIe, a Raspberry Pi 5 (test) is required to use the AI Kit.
Get started right away thanks to ready-made models
The AI Kit should be usable directly, even without previous experience. The AI accelerator can be integrated as a postprocessor in the official camera software, Picam. Raspberry Pi Limited and Hailo have developed some demo applications together, and you can try them out right away.
In addition, Hailo provides a number of pre-trained models. You can then use this to create your own applications. The manufacturer’s Tappas library is used for this, which receives video data via a Gstreamer interface. More experienced users can also bring their own models to the accelerator using Hailo’s software environment. These can be implemented using the Tensorflow (Lite), Pytorch, ONNX, or Keras frameworks.
Price and availability
The Raspberry Pi Limited is calling for 70 US dollars for the AI Kit, which is currently equivalent to around 65 euros. The kit is currently not available from German retailers, but Pimoroni lists it as available for pre-order. This makes it significantly more expensive than Google’s Coral in M.2 format, which costs $40 with two Edge TPUS. However, together, they only have 8 TOPS and require 4 watts of electrical power. In comparison, the AI Kit stands up well. A camera is also required for image recognition; it does not necessarily have to be an official model.
Also Read: Artificial Intelligence: These are the professions at risk due to AI technology
